KR100714515B1 - Method and elevator scheduler for scheduling plurality of cars of elevator system in building - Google Patents

Abstract

The present invention relates to a method of scheduling a car of an elevator system in a building. This method is initiated whenever a newly arrived passenger presses the up or down button to create a call to the service. Based on the future situation of the elevator, once a car is assigned to service a call, for each car a first waiting time for all current passengers is determined.

Description

Elevator scheduler and method for scheduling a plurality of cars of an elevator system in a building TECHNICAL TECHNICAL TECHNICAL FIELD OF ELEVATOR SCHEDULER FOR SCHEDULING PLURALITY OF CARS OF ELEVATOR SYSTEM IN BUILDING

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention generally relates to scheduling of elevator cars, and more particularly to an elevator scheduling method that takes into account future passengers.

Elevator scheduling in large buildings is a difficult industrial problem. This problem is characterized by very large state space and significant uncertainty, referring to Barney's "Elevator Traffic Handbook" (Spon Press, London, 2003). Typically, a passenger requests elevator service by pressing a call button. This allows the elevator scheduler to assign an elevator car to service the passenger.

Early elevator schedulers used the principle of collective group control. In this heuristic, refer to Strakosch's "Vertical transportation: elevators and escalators" (John Willy & Sons, 1998). Assign a nearby car to service the passenger. Such scheduling is not appropriate and unpredictable. For this reason, this collective control is difficult to accept when passengers expect to be notified that the car will pick them up, immediately after the call.

Another discovery is to minimize a remaining response time (RRT) for each passenger. This RRT is based on the current schedule, referring to US Patent 5,146,053, "Elevator dispatching based on remaining response time," registered on September 8, 1992 by Powell et al. It is defined as the time taken for the pickup of each pre-defined passenger. These findings focus solely on minimizing passenger wait times, completely ignoring the impact of current assignments on future passenger wait times.

Among the minimizations based on RRT, "Elevator dispatchers for down-peak traffic" by Bao (Massachusetts, Amherst, University of Massachusetts, Department of Electrical and Computer Engineering, Technical Report) (1994), and how to ignore the uncertainties associated with passenger demand floors and, as a reference, by "Decision-theoretic group elevator scheduling" by Nikovski et al. US Patent Application No. 10 / 161,304, filed June 3, 2003 by Brand, et al. (Int'l, Transto, 13th International Conference on Automation Planning and Scheduling, June 2003) and Brand et al. "Method and System for Dynamic Programming of Elevators for Optimal Group Elevator Control" When referenced, it may be further distinguished by methods for properly determining the expected RRT of each passenger for the destination floor.

However, the uncertainty associated with future passengers is a brand new challenge for at least two reasons. It is a complex matter to properly consider the impact of current decisions regarding the waiting times of all future passengers. First, the uncertainty associated with future passengers is very high because they do not know all the arrival time, arrival floor and destination floor. Second, the current decision may potentially affect passenger waiting time arbitrarily until the distant future, which creates a theoretical optimal range for the infinite problem.

Despite computational difficulties, ignoring future passengers often results in less than optimal scheduling results. This current assignment will affect the future operation of the car, affecting the ability to service future calls with a minimum amount of time.

One particular situation that illustrates the importance of future passengers is peak traffic. During down-peak traffic periods, for example during or after work on weekdays, most future passengers choose the main floor as their destination. Since these future passengers are largely distributed across the upper floors, scheduling for down-peak traffic is a very difficult problem.

During up-peak traffic periods, most prospective passengers arrive at the main floor and request service to the upper floor. Typically, the up-peak period is much shorter, more frequent and concentrated than the down-peak period. Thus, up-peak throughput is typically a limiting factor that determines whether an elevator system is suitable for a building. Accordingly, it is important to optimize the scheduling process for up-peak.

Consider the following scenario. At some level the call is made. Only one car is parked on the main floor, and the scheduler decides to only service the call based on the passenger's expected waiting time. If a car at the main floor is dispatched to service a call, the main floor will be free of cars, so future passengers will have to wait longer than when the car is waiting. This shortsighted decision, which is commonly found in conventional schedulers, is particularly difficult during up-peak traffic, because the main floor is filled with many waiting passengers in a short time while the car serves one of the passengers mentioned above. do.

Some elevator scheduling methods are known to consider future passengers according to changing future outcomes. Some schedulers, "The revolutionary AI-2000 elevator group-control system and the new intelligent option series" by Ujihara et al. (Mitsubishi Electric Advanced, 45: 5 8, 1998), using fuzzy rules that identify situations similar to those described above, make more sensitive decisions about future work. However, this method has a significant disadvantage. First, this rule needs to be coded manually. Thus, the system is only useful for "experts". Second, the interpretation of fuzzy rule inference between rules is often erratic, especially when there are no applicable rules for any particular situation. Thus, elevators often operate in unintended and erratic ways.

Another approach recognizes that group elevator scheduling is a series of decisions that cause problems. This method is described in Crites et al, "Elevator group control using multiple reinforcement learning agents" (Machine Learning, 33: 235, 1998). It uses a Q-learning algorithm that asynchronously updates all future states of the elevator system. They handle the massive state space of the system by neural networks, which approximates the cost of all future states. These approaches show important possibilities. However, its computational requirements make it completely impractical in commercial systems. The elevator operation experiment for this method, concentrated on a single traffic profile, takes approximately 60,000 hours, and the resulting reduction in latency for other faster algorithms is only 2.65%, which does not justify its computational cost.

Conventional methods are either strenuous running or computationally expensive, or both. Therefore, there is a need for a method of optimally scheduling an elevator car, considering future passengers, especially for up-peak traffic intervals.

Disclosure of the Invention

The present invention provides a method of scheduling a plurality of cars of an elevator system in a building. The method includes receiving a call and, once a car is assigned to service the call, determining, for each car, a first waiting time for all current passengers, based on the future state of the elevator system. Determining a second waiting time for future passengers, for each car, for each car, based on the landing patterns of the plurality of cars, if the car is assigned to be serviced for the call. Combining the second wait time to produce an adjusted wait time and assigning a particular car with the lowest adjusted wait time to service the call and minimize the average wait time of all passengers.

1 is a block diagram of an elevator system used in the present invention,

2 is a flow chart of a method for scheduling an elevator car in accordance with the present invention;

3 is a grid showing Markov chains according to the invention.

Best Mode for Carrying Out the Invention

<System structure>

1 shows an elevator scheduler 200 according to the present invention in a building 104 having upper floors 102, a main floor 103, elevator shaft 104, and elevator car 105. ). The main floor is mainly the ground or lobby floor, ie the floor where most passengers enter the building.

For the purposes of the present invention, passengers are formally classified into several classes according to variables representing known information about the passengers. These variables introduce uncertainty into the elevator scheduler's decision process. These classes are boarding passengers, waiting passengers, new passengers and future passengers.

For each passenger 111, the arrival time, arrival floor and destination floor are all known. Passengers are in the car and no longer waiting.

For each waiting passenger 112, the arrival time, arrival floor and direction of travel are known. The target layer is unknown. Cars are assigned to service each waiting passenger.

As for the new passenger 113, since the new passenger sends the call signal 120, the arrival time, arrival floor, and driving direction are known. Assigning a car to service this new passenger's call is a common problem. Sometimes there is only one new passenger.

The above-described third class passengers 111 to 113 are commonly 'existing' passengers. In this specification, the reason for these passengers to be 'current' is that they have already arrived physically, and the system knows some information about all these passengers. Among the current passengers, only waiting passengers and new passengers have non-zero waiting times.

Nothing is known about future passengers 114 that do not yet exist. The best solution is to probabilistically represent passenger variables by random variables, or guess from historical data. All passengers include current and future passengers.

It is a specific problem to assign a car to service new passengers to minimize the expected waiting time for all passengers, ie current and future passengers.

<Operation method>

2 shows a method for scheduling a car of an elevator system 100 according to the invention. This method is executed in response to the call 201. This call is possible on all floors. First, the scheduler 200 assigns a car to service for the call, and for each car, based on the future state 209 of the elevator system, the first expected wait for all current passengers 111-113. Determine time 211. Second, the scheduler assigns the car to service for the call 102 and, for each car based on the landing pattern 219 of the car 105, for each car a second expected wait of future passengers 114. Determine time 221. For each car, the first and second expected wait times are combined 230 to generate an adjusted wait time 231 and to assign a car with the lowest adjusted wait time to service the call 201 ( 240).

Ideally, the elevator scheduler should, before assigning, integrate all sources of uncertainty to determine the marginal cost of all possible assignments. However, due to the insurmountable computational complexity of the scheduling problem, the majority of commercial elevator schedulers usually rely on discovery methods that ignore some or all of this uncertainty.

In typical up-peak traffic periods, a significant number of future passengers, such as 80% to 95%, arrive at the main floor. This waiting time for main floor arrival is a major factor in the total waiting time of the elevator system during the up-peak traffic period, and the current decision of the elevator scheduler is to minimize the expected waiting time of passengers on the main floor.

Thus, the present invention begins with the simple assumption that all future passengers arrive at the main floor. The effect of not modeling arrivals at other floors in the future results in a shorter time-horizon in which the expected atmosphere is accurate in the near future. However, this effect, like the reduction factor, is clearly included in subsequent calculations. In addition, during the up-peak traffic period, most future passengers actually arrive at the main floor.

Under this assumption, the current decision of the elevator scheduler affects future passenger wait times by the future arrival of the car to the main floor. In the present invention, the arrival sequence of the car to the main floor is called a landing pattern.

For the purposes of the present invention, the landing pattern 219 of the car to the main layer is determined by the following factors. First, passengers on the upper floor may select the main floor as their destination. Secondly, the empty car can automatically select the main floor as a place to park while waiting for the next call. By determining the landing pattern 219, the individual future passengers 214 are effectively ignored.

The optimum parking plan and its impact on the landing pattern are described in US Patent Application No. 10 / 293,520 filed November 13, 2002 by Brand et al., Described below, as "Optimal Parking of Empty Cars in Elevator Group Control." of Free Cars in Elevator Group Control).

One plan for servicing the main passengers is to send each car preferentially to the main floor immediately after completing the service for the last passenger. In a building with C cars, the landing pattern 219 is a time array

, 단 T_j≥0

, Where T _j ≥0

Where T _j (j = 1, ..., C) is the arrival time of the car to the main floor after transporting all the passengers.

Since there is uncertainty about the destination of the waiting passengers 112 and the new passengers 113, the landing pattern T is a vector value random variable with probability distribution P ( T ), and the space of all possible landing patterns T 219. T ∈ T for.

로서 그 시간을 예상한다.Ideally, the scheduler 200 determines the expected waiting time V ( T ) for each possible landing pattern T ∈ T , for the probability distribution P ( T ).

Expect that time as.

Where <> represents an expectation operator. Indeed, this is an accurate estimate of the waiting time of passengers in the main floor, under the assumption that all new passengers are at the main floor. However, there is no practical way to determine the probability distribution P ( T ). If any, the size of the space of all possible landing patterns is very large. Integrating over this space is computationally unrealistic.

을 사용하며, 근사값

을 사용한다. 각각의 성분 T_j(j=1, …, C)가, 카 j에 배정된 탑승 승객 및 대기 승객들의 목적지에 대한 확률 분포에 따라 달라지는 불확실성을 갖는 독립 랜덤 변수이기 때문에,

가 사실이라는 것을 주의해야 한다.Instead, the present invention includes alternative landing patterns that include individual estimated arrival times at each car's main floor.

, Approximation

Use Since each component T _j (j = 1,…, C) is an independent random variable with uncertainty that depends on the probability distributions for the destinations of the boarding passengers and waiting passengers assigned to car j,

Note that is true.

은, 물론, 현재의 승객들에게로의 좀더 빠른 배정 및 그 불확실한 목적지에 따라 달라진다. 즉, 랜딩 패턴은 현재의 승객들(111~113)의 예상 대기 시간(211)에 따라 간접적으로 달라진다. 현재의 승객들(111~114)의 예상 대기 시간(211)을 결정(240)하는 방법은, 이하의 참조로서 이용하는, 니코프스키 등에 의한 2003년 6월, 자동화 프래닝 및 스케줄링에 대한 13회 국제 회의의 "결정-이론 그룹 엘리베이터 스케줄링", 및 브랜드 등에 의해 2002년 6월 3일에 제출된 미국 특허 출원 10/161,304, "최적의 그룹 엘리베이터 제어를 위한 엘리베이터의 다이내믹 프로그램용 시스템 및 방법"에 기재되어 있다. 짧게 말하면, 이 방법은 "다이 내믹 프로그램에 의한 시스템 알고리즘 비우기"(ESA-DP) 방법으로 불리워진다.For the same reason, this approximation is well suited for averaging. Accurate landing time for each car

This, of course, depends on faster assignments to current passengers and their uncertain destinations. That is, the landing pattern is indirectly changed according to the expected waiting time 211 of the current passengers 111 to 113. The method of determining 240 the expected waiting time 211 of the current passengers 111-114 is 13 times for automated planning and scheduling, June 2003 by Nikowski, et al. "Decision-Theory Group Elevator Scheduling" at the International Conference, and US Patent Application No. 10 / 161,304, filed on June 3, 2002 by Brand et al., "Systems and Methods for Dynamic Programming of Elevators for Optimal Group Elevator Control". It is described. In short, this method is called the "empty system algorithm by dynamic program" (ESA-DP) method.

을 고려하였다. 그러나, 스케줄러(200)의 현재 결정, 즉 신규 승객(113)에 대해 서비스하도록 카를 배정하는 것은 이러한 배정을 변경한다. C개의 카 중 어느 하나를 스케줄러가 선택할 수 있기 때문에, C개의 가능한 최종 배정 및, 그에 따른 C개의 가능한 랜딩 패턴(219)의 분포가 존재한다. 본 발명에서 전술한 근사값을 이용하면, 본 발명에서는 카 I에 신규 승객(113)이 배정될 때에 발생하는 랜딩 패턴So far, the present invention relates to a parking pattern T as a function of the assignment of fixed current passengers to the car.

Considered. However, the current decision of scheduler 200, ie, assigning a car to service new passengers 113, changes this assignment. Since any one of the C cars can be selected by the scheduler, there are C possible final assignments and hence the distribution of C possible landing patterns 219. Using the above approximations in the present invention, in the present invention, a landing pattern that occurs when a new passenger 113 is assigned to the car I.

(단, I=1, …, C)

(Where I = 1,…, C)

의 평균은 신규 승객(113)이 카 I에 배정될 때에 카 j의 예상 랜딩 시간이다.This is necessary. Each entry

The average of is the estimated landing time of car j when new passenger 113 is assigned to car I.

After the matrix for landing patterns 219 for the C cars has been established, the expected cumulative waiting time 221 of the future passengers 214 corresponding to each landing pattern, i.e., the rows of the matrix, can be determined. have. The present invention provides a procedure for determining the expected wait time of future passengers 214 as a function of any landing pattern 219.

T = [T ₁ , T ₂ ,... , T _c ]

(여기서, n(t)는 시간 t에서 주층(103)에서 대기하는 승객들의 예상 수임) 이내에 모든 장래 승객들(114)의 예상 누적 대기 시간(221)으로서 V⁰(T)를 규정한다.Since the estimated time 221 of future passengers 214 is invariant for the specific order of arrival of the car, that is, whether the car "2" has arrived within 10 seconds and vice versa whether the car "3" has arrived within 15 seconds. Is not different. In the present invention classifies the landing pattern T (219) in ascending order _{(0≤T 1 ≤T 2 ≤ ... ≤T} c). With this assumption, in the present invention the time interval

N (t) defines V ⁰ (T) as the estimated cumulative waiting time 221 of all future passengers 114 within (where n (t) is the expected number of passengers waiting at main floor 103 at time t).

Before describing the car assignment procedure, the present invention introduces an exponential decrease in future wait time 221 due to the bias within the predicted parking time of the car. This bias is due to an approximation that there is no future arrival on the main floor before the end of the current landing pattern.

Indeed, even if this future arrival is not frequent. These passengers will be assigned to cars with boarding passengers and waiting passengers. These cars are then delayed upon arrival at the main floor. Thus, the landing time estimated by ESA-DP processing may slightly estimate the actual time for near future predictions, and perhaps considerably less for far future predictions.

The near future may be defined as the average time it takes for the car to return from the main floor, for example 40 to 60 seconds in a medium sized building. This time can be calculated.

One way to reduce the distant future estimate is to multiply that estimate by exp (−βt), where β> 0 is the decreasing factor.

로 규정한다. 간격 [0, T_c]는 C개의 다른 간격 [T_i-1, T_i](i=1, …, C]로 분할되고, T₀=0으로 설정한다. 시간 t∈[T_i-1, T_i]에서 대기하는 승객들의 예상 수는 주층에서 랜딩된 카의 최종 시간이 (T_i-1)인 이후에 경과하는 시간에 비례한다.As in the case described above, the present invention provides for the expected reduced cumulative latency of future passengers.

It is prescribed by. The interval [0, T _c ] is divided into C different intervals [T _i-1 , T _i ] (i = 1,…, C] and sets T ₀ = 0. Time t∈ [T _i-1 , The expected number of passengers waiting at T _i ] is proportional to the time that has elapsed since the last time (T _i-1 ) of the landing car on the main floor.

이고, 전술한 적분은 값을 구할 수 있는 C개의 부분으로 나누어진다. 본 발명에서는 로딩 시간이 대기 시간에 비해 작기 때문에, 카가 주층에서 대기하는 모든 승객들을 즉시 픽업할 수 있다고 가정한다.In the present invention, when the arrival of future passengers 114 is modeled as a Poisson process with a ratio λ, the expected number of passengers at the main floor is

The above-described integral is divided into C parts from which values can be obtained. In the present invention, since the loading time is smaller than the waiting time, it is assumed that the car can immediately pick up all the passengers waiting on the main floor.

However, if the car i arrives at the main floor and judges that there is no one, it does not leave immediately at its arrival time Ti. Instead, the car waits at the main floor until future passengers signal 120 and switch to new passengers 113. If there are j cars on the main floor at time t = 0, the first j passengers do not wait at all. Each passenger boards the car immediately and has no waiting time. Sufficient but speculative savings in this scenario are balanced against the actual cost of not using the car to service new passengers on the upper floor. In order to measure the amount of these savings, the elevator car on the main floor is accurately modeled.

<Semi-Markov Model>

In order to accurately estimate the waiting time 221 of future passengers 214, assuming the actual operation of the car when there is no waiter on the main floor, in the present invention, the state and the transition of the cars landing on the main floor in the present invention. Use a semi-Markov chain to describe

, 각 쌍의 상태 S_i와 S_j간의 전이의 가능성 P_ij, 초기의 분산

를 포함한다. 또한, 각 세미-마르코프 체인은 이산 시간으로 전개되는 엠베드된 풀리-마르코프 체인(embeded fully-Markov chain)을 포함하고 있으며, 누적 전이 비용 R_ij가

로서 규정되고, 모든 전이가 시간 단위 내에서 발생하도록 가정하고 있다. 본 문제를 위해 이용하는 세미-마르코프 체인에서의 상태는 (i, j, m)에 의해 분류되며, 여기서 I는 주층에서 아직 랜딩되지 않은 카의 수이고, j는 승객들을 위해 대기하는 주층에서의 현재 파킹된 카의 수이며, m(=C-i-j)은 주층으로부터 벌써 출발한 카의 수이다.Semi-Markov chains are described in Bertsekas's "Dynamic Programming and Optical Control" (Belmont, Mass., Athena Scientific Volumes 2, pp. 261-264, 2000). Finite number of states S _i (i = 1,…, Ns), average instantaneous cost i _ij , expected transition time, specifying the possibility of starting in this state S _i

, The probability of transition between each pair of states S _i and S _j P _ij , the initial variance

It includes. In addition, each semi-Markov chain contains an embedded fully-Markov chain that is deployed in discrete time, with a cumulative transition cost R _ij .

And assume that all transitions occur within a time unit. The states in the semi-Markov chain used for this problem are classified by (i, j, m), where I is the number of cars not yet landed on the main floor, and j is the current on the main floor waiting for passengers. The number of cars parked, m (= Cij) is the number of cars already departed from the main floor.

As shown in Fig. 3, the present invention organizes the states of the semi-Markov chain in a two-dimensional grid or matrix. Each element S _im 301 in the matrix 300 corresponds to a state (i, j, m). The grid structure in FIG. 3 is for a semi-Markov chain embedded in a building with four shafts. Row I of model 302 includes all possible states of the system immediately after car I picked up all passengers waiting at the main floor arriving at time T _i . Note that the vertical time axis 303 is not drawn to scale. Only the transition shown by the bold arrow costs money. The cost of all other transitions is zero. A transition 305, designated n + for any number n, occurs when more than n passengers arrive.

First, the present invention provides a solution to the general situation represented by this model, i.e. when there is no car parked at the main floor at the current decision time (T ₁ > 0), and there are several cars parked at the main floor thereafter. The solution is magnified.

In a typical situation, the starting state of the chain is state (C, 0, 0), ie all C cars have not yet landed on the main floor. The final state is the state of the bottom row of the model when all C cars are landed and depends on how many future passengers arrived within the interval t∈ [0, T _c ]. When passengers board and all cars depart, i.e. state (0, 0, C) 210, or a few cars still remain on the main floor, i.e. state (0, j, Cj) (about j> 0) any one.

에 의해 쉽게 결정될 수 있다. 각각의 전이 확률도, 도착율 λ의 푸아송 처리로부터, 고정 간격 내에 특정 수의 장래의 승객들이 도착할 확률과 동일하기 때문에 결정될 수 있다. 따라서, x명의 승객들이 시간

내에 정확히 도착할 확률 p(x)는

이다. 도착하는 승객들의 정확한 수로 명기된 전이의 경우, 이 식을 직접 사용할 수 있다. n=라고 명기된 전이는 n명 이상의 신규 승객들이 도착할 때 이루어지는 것을 의미하고, 전이 가능성은 이 상태로부터 모든 남아있는 출발하는 전이의 확률의 합계를 1에서 뺀 것이다(

).Each state (i, j, m) (where j = Cim) in the row above one (i> 0) of the bottom row may transition to two or more subsequent states. This depends strictly on how many future passengers arrive during the time interval t∈ [T _i , T _{i + 1} ]. For example, the chain transitions from state (4, 0, 0) to state (3, 1, 0) only when there are no passengers arriving by time T ₁ , and state (3) when more than one passenger arrives by that time. , 0, 1). Each transition in FIG. 3 is specified by the number of passengers arriving when this transition is made. The time to complete each transition is the interval between when two cars arrive

Can be easily determined. Each transition probability may also be determined from the Poisson process of arrival rate [lambda] because it is equal to the probability that a certain number of future passengers arrive within a fixed interval. Thus, x passengers time

The probability p (x) that arrives exactly within

to be. For transitions specified by the exact number of passengers arriving, this equation can be used directly. Transitions denoted by n = means that when more than n new passengers arrive, the transition probability is the sum of the probabilities of all remaining departures from this state minus one (

).

Determining the cost of the specified number of passengers is made immediately because the number of arriving passengers is less than or equal to the number of cars parked on the main floor. These passengers do not wait, so the cost of the corresponding transition is zero. However, it is very relevant to determine the cost of most transitions of the final or right side from each state. Such a transition corresponds to the case where more than n passengers arrive at the main floor while n-1 cars are parked at the main floor. This calculation must account for the fact that when x future passengers arrive (x≤n), first n-1 passengers board the car and leave without waiting, and the remaining x-n + 1 passengers must wait. do.

FIG. 3 shows that for any state S _im of the grid, the transition shown in bold lines is made when j or more future passengers arrive, i.e., when n = j-1, as described above (j = Cim). Indicates. Thus, when x transitions have arrived and x future passengers arrive, only the last xj passengers should wait. That is, if x passengers appear within a certain time t, the derivative or instantaneous cost r _im at that time is xj.

Since this transition has a certain number of passengers of j or more, and even within a finite time interval, this number can be arbitrarily large in theory, so the estimated cost of the transition is from j + 1 to infinity, weighting all possible arrivals x or more. Sum, this weight is the probability that x arrivals will occur, as determined by the Poisson distribution.

In addition, the derivative cost at time t can be reduced by the coefficient of exp (−βt), as described above. This inference yields the following expression for the expected reduced cumulative waiting time Rβ _im of the main passengers during the final transition from state S _im (j = Cim).

Depending on the two components of the difference between x-j, after changing, simplifying and dividing the integral into two parts, the representation of the cost is a function of

로 산출되지만, 임의의 일정한 적분 상수 C₀은 본 발명에서 편의상 0으로 설정한다.By using

Although, the constant constant constant C ₀ is set to 0 for convenience in the present invention.

After determining all the costs and probabilities of the semi-Markov model as described above, the cumulative cost of the atmosphere generated by the system when launched in any model state is described by Bertsekas in "Dynamic Programming and Optimal Control." (Dynamic Programming and Optical Control) ”(Belmont Athena Scientific Volume 1, pages 18–24, 2000), dynamic programming can be effectively determined by starting from the bottom row of the model. Since the state in the bottom row is final and indicates the end of the landing pattern, the present invention sets the waiting time to 0, i.e., it is not interested in the amount of waiting time accumulated since the last landing.

After determining the latency for all states, the present invention can obtain the cumulative latency for the entire pattern T from the infinite state of the model. In the general case, if there is no car in the main floor at time t = 0, the infinite state is always (C, 0, 0). In certain cases, when one or more cars are parked on the main floor at time t = 0, they can be easily and readily manipulated. In this particular case, the starting state is (Cl, l, 0), where l is the number of cars in the main floor, and the expected reduced cumulative wait for the entire pattern is the waiting time (S _C-1 , 0) of this starting state. )to be. This eliminates the need to handle this particular case separately from the general case.

을 제공한다.The above-described procedure is based on the respective landing pattern T _i 219 due to the decision to assign the current call 201 to car i (i = 1, ..., C), to predict future passengers 114. Equation of cumulative decrease latency (221)

To provide.

At the same time, the ESA-DP processing includes the current passengers, including the new passenger signaling the call 210, at step 210 when the call is assigned to car i (i = 1,…, C) (230). The estimated value Wi of the cumulative non-reduction wait time 211 of (211 to 213) is determined.

In order to reach an optimal decision to balance the current passenger's wait 211 and the future passenger's wait 211, the setting of the two values V _i β and W _i are combined to determine the adjusted waiting time 231. 230).

There is a significant difference between these two measurements. The cumulative waiting time W _i 211 of the passengers, namely the waiting passenger 112 and the new passenger 213, is not reduced, while the cumulative waiting time 221 of the future passengers 214 is reduced.

Moreover, the purpose of the scheduling process 200 is to minimize the average wait time, not to minimize the cumulative wait time for a given interval. For optimization, two measurements can only be interchanged if the time intervals for all possible decisions are the same.

In general, this is not the case. The landing pattern does not have the same duration for each car. Thus, scheduling process 200 should average the wait times from their counterparts.

(211)을 얻는 것은 간단하다. 현재의 승객들(11~113)의 수 N은 스케줄러에 의해 항상 알려져 있으며, 후보 카 수 I에 의존하지 않아,

이다. 한편, 랜딩 패턴(219)의 지속 기간에 대해 누적 감소 대기 시간 V_iβ으로부터 장래의 승객들

(214)의 평균 대기 시간(221)을 얻는 것은 명백하지 않다.Average estimated waiting time of current passengers _11-113 from cumulative waiting time W _i

Getting 211 is simple. The number N of current passengers 11-113 is always known by the scheduler and does not depend on the candidate car number I,

to be. On the other hand, future passengers from the cumulative decrease latency V _i β for the duration of the landing pattern 219

It is not clear to obtain an average wait time 221 of 214.

The period T _c of the landing pattern is known. If the arrival rate of the main floor is λ, then the expected number of arrivals within the T _c time unit is λ T _C. However, dividing V _i by λ T _C is meaningless because V _i decreases at a decreasing rate β.

는 시간 간격 [0, T_c] 동안에 도착하는 승객들의 예상 누적 가중 수를 의미한다. 따라서, 양

은 이 간격 이내에 도착하는 장래의 승객들의 예상 평균 수이며, 모든 가중 계수의 적분 합에 의해 적절히 표준화된다. 더욱이, 리틀의 법칙(Little's law)은, 카산드라(Cassandras) 등에 의한, "이산 이벤트 시스템의 소개(Introduction to discrete event systems)"(네덜란드 도르드레츠, 크루베르 아카데믹 퍼블리셔, 1999년)를 참조하면,

임을 특정한 다. 이는 최종적으로 장래 승객들의 시간 표준화된 예상 대기

를 산출한다.Instead, the reduction factor exp (-βt) is the averaging weight for time t. As reflected in the Markov model's cost, if n (t) is the expected instantaneous number of passengers arriving at time t,

Is the expected cumulative weighting number of passengers arriving during the time interval [0, T _c ]. Thus, the amount

Is the expected average number of future passengers arriving within this interval and is properly standardized by the integral sum of all weighting factors. Furthermore, Little's law refers to Cassandras et al. "Introduction to discrete event systems" (Netherlands Dordrez, Krueber Academic Publishers, 1999).

Is specified. This is finally the time standardized expected wait for future passengers

Calculate

(211) 및

(221)를 얻음으로써, 이들 대기 시간은, 예컨대 가중치에 의해서(0≤α≤1), 단일의 조정된 대기 시간(2231)으로 조합되어(230), 조정된 대기 시간은

으로 된다.Computable estimate of waiting time for current and future passengers

(211) and

By obtaining (221), these waiting times are combined (230) by a single adjusted wait time (2231), for example by weight (0 ≦ α ≦ 1).

Becomes

The balance between the current passenger and the future passenger's wait depends on how quickly the system is freed from the current limitations by transporting passengers.

Thus, the optimal value of α can be empirically determined based on the physical operating characteristics of the elevator system. In the present invention, regardless of the height of the building and the number of shafts, it was found that weights within the interval [0.1, 0.3] yield a stably acceptable result.

(Effects of the Invention)

The system and method as described above can considerably reduce the waiting time by saving about 5% to 55% of the waiting time for the conventional scheduling process. Their improvement is due to the anticipation policy for future passengers. The performance of an elevator in up-peak traffic typically determines the number of shafts needed in the building. The present invention can often reduce the number of shafts required for mid- and high-rise office buildings, one by one, while providing excellent service by using standard guidelines for installing elevators in buildings. In a medium story building, such as a 25-30 story building, the cost per elevator can be about $ 200,000. The removal of the shaft not only reduces the cost of the building, but also reduces maintenance costs and increases the available floor space.

While the invention has been described in terms of preferred embodiments, it will be understood that various other applications and modifications are possible within the scope and spirit of the invention. Accordingly, the objects of the appended claims, which come within the scope and spirit of the invention, include all such modifications and variations.

Claims (15)

A method for scheduling a plurality of cars of an elevator system in a building, Receiving a call, When assigning a car to service the call, determining, for each car, a first waiting time for all current passengers, based on the future state of the elevator system; When assigning a car to service the call, determining, for each car, a second waiting time of future passengers, based on the landing patterns of the plurality of cars; Combining the first and second waiting times to produce, for each car, an adjusted waiting time; Assigning a particular car with a lowest latency adjusted to service the call and to minimize the average latency of all passengers, The landing pattern is determined for a near future time interval, The near future time interval is an average time taken to return from the main floor of the plurality of Kaga buildings. A plurality of car scheduling methods of an elevator system. The method of claim 1, The current passengers are boarding passengers in the plurality of specific cars knowing arrival time, arrival floor and destination floor, waiting passengers assigned to the plurality of cars knowing arrival time, arrival floor and direction of travel, and A method of scheduling a plurality of cars in an elevator system, including new passengers sending a call signal, wherein all passengers include current and future passengers. The method of claim 1, The determining of the first waiting time may include: Calculating a cost function that determines the cost for each future state, Assigning a particular car associated with the series of states with the lowest cost; A scheduling method of a plurality of cars in an elevator system. The method of claim 1, A method of scheduling a plurality of cars of an elevator system in which a considerable number of said future passengers arrive at a selected floor during up-peak traffic periods. The method of claim 1, The landing pattern of the car of the elevator in the selected floor is a random variable T of vector values having a probability distribution P ( T ), and T ∈ T for the space of all possible landing patterns T. The method of claim 5, And all possible landing patterns depend on the landing times of the plurality of cars. delete delete The method of claim 1, And wherein said landing pattern for a distant future time interval t is reduced by exp (-[beta] t) and [beta]> 0 is a reduction factor. The method of claim 4, wherein A method of scheduling a plurality of cars of an elevator system in which future passengers arrive at the main floor according to a Poisson process with a ratio λ. The method of claim 1, And said landing pattern is modeled by a Semi-Markov Chain having a plurality of states and transitions. The method of claim 1, Wherein the first waiting time W and the second waiting time V are combined according to αW + (1-α) V, where α is a weight within a range of 0 ≦ α ≦ 1. The method of claim 12, Optimal weight α is a scheduling method for a plurality of cars in an elevator system within an interval [0.1, 0.3]. The method according to claim 4 or 5, Wherein said selected floor is the main floor of said building. An elevator scheduler for scheduling a plurality of cars of an elevator system in a building, Means for receiving a call, Means for determining, for each car, a first waiting time for all current passengers, based on the future state of the elevator, when assigning the car to service the call; Means for determining, for each car, a second waiting time of future passengers, based on the landing patterns of the plurality of cars, when assigning the car to service the call; Means for combining the first and second waiting times to produce an adjusted waiting time for each car, Means for servicing the call and assigning a particular car with a lowest latency adjusted to minimize the average latency of all passengers, The landing pattern is determined for a near future time interval, The near future time interval is an elevator scheduler that is the average time taken to return from the main floor of the plurality of Kaga building.

Images